Much effort is currently directed towards the design of a responsive (closed-loop) stimulation device for the treatment of epilepsy (Osorio 2001, Sun 2008).The ideal intervention system would be able to prevent the occurrence of a seizure before the onset of behavioral and clinical symptoms, so as to impose minimum cognitive and emotional effects on the patient. However, current published seizure prediction algorithms are still too limited to be incorporated into these closed-loop devices. The aim of this project is to develop a sensitive and specific seizure prediction algorithm. We propose to use novel measurements of brain dynamics to improve the performance of current algorithms. Clinical and laboratory findings support the existence of a 'preseizure' state. In theory, such a state should be detectable. Despite initial promising results, detecting the preseizure state has proven to be a challenging task, in part due to the confounding effect of state of vigilance (Wake, non-rapid eye movement sleep, rapid eye movement sleep) on seizure dynamics (Schelter 2006). In reviewing the state-of-the-art in seizure prediction in 2007, Mormann et al. (Mormann 2007) used careful statistical analysis of published results to document that despite some evidence of success, seizure prediction was still quite difficult and they attributed a failure to account for SOV as one of several problems. Despite these warnings, little effort has been put forth to incorporate SOV in published seizure prediction algorithms. Furthermore, most seizure-prediction algorithms utilize passive measurements of brain dynamics, such as spontaneous EEG. We propose here to incorporate SOV as well as active measurements of brain dynamics as additional feature types to be used as input to the seizure prediction algorithm: We will probe the brain with polarizing low-frequency electric field (PLEF) stimulation and record the response during the preseizure (period immediately before seizure onset) and interictal (period well before seizure onset, between two seizures) periods. We will extract several characterizing feature sets from these responses and use them to distinguish the preseizure state. We will conduct these experiments in the rat tetanus toxin model of temporal lobe epilepsy. In SA1, we will develop a seizure predictor based on several different features extracted from the spontaneous EEG, including the state of vigilance. We will use this passive predictor as our working baseline. In SA2, we will implement an active probing paradigm to monitor the dynamic state of the brain by recording neural responses to PLEF between and before seizures. We will use these responses, along with SOV, as input features to augment our predictive algorithm. In SA3, we will use computational modeling to gain insight to the underlying dynamics of sleep and the intimate link between sleep state and seizure (Dinner 2002). Wewill use algorithms developed under this aim to reconstruct 'hidden' variables in a dynamical model of sleep. These variables will be evaluated as potential input features for our seizure prediction algorithm.